Modernize the work of the CFO: data, AI and decisions
How the role of the CFO is evolving in 2026. From administrative management to financial strategy: digitalization, AI, ESG, leadership. Key skills of the modern CFO.

Co-founder and CFO, Iter Advisors

Remember in 30 seconds
Modernizing finance starts with controlled data, defined responsibilities and testing on a limited flow. The gains are measured on the file; AI does not replace the review of sources and results.
The role of the CFO evolves when the company changes its data sources, its tools or its organization. The priority remains the same: to make the figures reliable and explain their consequences for decisions. For position responsibilities, see the profession of CFO. This guide addresses developments in practices, without presuming a saving of time or a universal distribution of work.
1. Evaluate what is changing in financial work
A closing, a budget and a management meeting do not use the same data. Before modernizing the function, observe how information is produced, corrected and validated. Re-entries, missing parts and unexplained discrepancies are different problems. They are not all solved by software.
It is not possible to deduct the time freed up by digitalization from the job title. A useful measure relates to a defined task, over a comparable period, including corrections and review. A result obtained in a company does not automatically describe the market.
Management must also specify what remains in the team and what can be entrusted to a stakeholder. The guide Fractional CFO or interim distinguishes between mission formats. Changing the mode of intervention does not exempt from assigning decisions and validation powers.
2. Linking data, arbitrations and responsibilities
The first project is the definition of the data: periods, sources, calculation rules and responsible parties. A margin indicator is only useful if you know what expenses it covers. A cash balance does not describe expenses already incurred. The CFO organizes this reading with people who know the operations.
The second project is the preparation of arbitrations. The reporting must make it possible to discuss recruitment, an investment, an expense or a financing deadline. Present the assumptions and their effects on cash, results and schedule. A forecast is used to discuss scenarios; it does not guarantee their achievement.
The third project is the allocation of responsibilities. Producing a file, verifying an entry, presenting an analysis and authorizing a payment are distinct actions. Define access, controls and authorized people. The management control provides a reading of the discrepancies, while the decision rests with the agreed responsible parties.
3. Digitize a flow before extending the system
Choose a limited flow: collection of documents, reconciliation, updating of a forecast or distribution of a report. Compare the data produced with a reference source. Also test for a missing part, a duplicate and a period correction. These exceptions allow you to see if the team can actually exploit the system.
In Power BI, refreshing depends on the sources and the model used. The Microsoft documentation on data refresh explains these mechanisms. For financial management, above all specify the date of the last update and the action to take in the event of failure.
The guide choose and integrate financial tools presents the selection controls. Thedirectory of financial tools supplements this method with criteria specific to the products. A demonstration must cover your file and the functions included in the quote.
4. Supervising AI in financial production
An assistant can help prepare a summary, classify information or offer an initial analysis. The result must be able to be reconciled with the sources. Calculations, dates and references must remain verifiable; a convincing formulation is not proof of accuracy.
Start with limited use with authorized data and a review manager. Define what can be transmitted to the supplier, the storage conditions and access. THE questions and answers from the CNIL on generative AI help examine personal data risks. The choices must be validated according to the context of the organization.
The guide AI and automation of financial management distinguishes between uses and controls. To measure a pilot, note the preparation time, corrections and review on a constant perimeter. Only announce a customer result with the period, the measurement method and its publication agreement.
5. Prepare relevant extra-financial data
Requests from customers, financiers or partners may relate to extra-financial data. First identify the requests that are really applicable to the company, with the competent contacts. A list of indicators found online is not enough to define an obligation or a calculation method.
For each data retained, document the scope, the source and the person who validates it. Separate measured information, estimates and unavailable items. Keep supporting documents and explain method changes between two periods. The CFO can coordinate this work without replacing technical or legal specialists.
6. Make reporting a decision support
A financial meeting should result in a few decisions attributed to managers. Present the deviations that require action, their known causes and the hypotheses that are still uncertain. Agree on the next review and what information to obtain before deciding.
Useful communication distinguishes between what is done, what is forecast and what is recommended. An increase in income alone does not explain the change in cash. Financing does not constitute an improvement in profitability. The analysis must connect these dimensions without transforming a correlation into a mission result.
To reorganize, review the responsibilities and schedule in organize your financial management. A Fractional CFO can contribute to this work within an agreed scope. The data, the available team and the expected decisions determine the mission.
Sources consulted on October 2, 2026: Microsoft Learn and CNIL. The control examples are a proposed working method, not results of Iter missions or market statistics.